[Attention] Remove imports from vllm/attention/__init__.py (#29342)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
@@ -9,8 +9,9 @@ from tests.compile.backend import LazyInitPass, TestBackend
|
||||
from tests.utils import flat_product
|
||||
from tests.v1.attention.utils import BatchSpec, create_common_attn_metadata
|
||||
from vllm._custom_ops import cutlass_scaled_fp4_mm, scaled_fp4_quant
|
||||
from vllm.attention import Attention, AttentionMetadata
|
||||
from vllm.attention.backends.abstract import AttentionMetadata
|
||||
from vllm.attention.backends.registry import AttentionBackendEnum
|
||||
from vllm.attention.layer import Attention
|
||||
from vllm.attention.selector import global_force_attn_backend_context_manager
|
||||
from vllm.compilation.fusion_attn import ATTN_OP, AttnFusionPass
|
||||
from vllm.compilation.fx_utils import find_op_nodes
|
||||
|
||||
@@ -5,7 +5,8 @@ import pytest
|
||||
import torch
|
||||
|
||||
from tests.compile.backend import TestBackend
|
||||
from vllm.attention import Attention, AttentionType
|
||||
from vllm.attention.backends.abstract import AttentionType
|
||||
from vllm.attention.layer import Attention
|
||||
from vllm.compilation.matcher_utils import FLASHINFER_ROTARY_OP, RMS_OP, ROTARY_OP
|
||||
from vllm.compilation.noop_elimination import NoOpEliminationPass
|
||||
from vllm.compilation.post_cleanup import PostCleanupPass
|
||||
|
||||
@@ -14,7 +14,7 @@ import torch
|
||||
from torch._prims_common import TensorLikeType
|
||||
|
||||
from tests.kernels.quant_utils import native_w8a8_block_matmul
|
||||
from vllm.attention import AttentionType
|
||||
from vllm.attention.backends.abstract import AttentionType
|
||||
from vllm.model_executor.layers.activation import SiluAndMul
|
||||
from vllm.model_executor.layers.fused_moe.utils import moe_kernel_quantize_input
|
||||
from vllm.utils import (
|
||||
|
||||
@@ -5,8 +5,8 @@ import numpy as np
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.attention import Attention
|
||||
from vllm.attention.backends.abstract import MultipleOf
|
||||
from vllm.attention.layer import Attention
|
||||
from vllm.config import (
|
||||
CacheConfig,
|
||||
ModelConfig,
|
||||
|
||||
@@ -7,7 +7,7 @@ from vllm.v1.worker.utils import bind_kv_cache
|
||||
|
||||
|
||||
def test_bind_kv_cache():
|
||||
from vllm.attention import Attention
|
||||
from vllm.attention.layer import Attention
|
||||
|
||||
ctx = {
|
||||
"layers.0.self_attn": Attention(32, 128, 0.1),
|
||||
@@ -35,7 +35,7 @@ def test_bind_kv_cache():
|
||||
|
||||
|
||||
def test_bind_kv_cache_non_attention():
|
||||
from vllm.attention import Attention
|
||||
from vllm.attention.layer import Attention
|
||||
|
||||
# example from Jamba PP=2
|
||||
ctx = {
|
||||
|
||||
Reference in New Issue
Block a user